Week 1
(5/19/25 - 5/23/25)
Introduction
RCR training
Searching for prior research
Prompting techniques
A small example of prompting techniques utilizing different LLMs with a dataset
Week 2
(5/26/25 - 5/30/25)
Observed Memorial Day
Presented week 1 findings and Objectives
Observed U.S. Air Force graduation, 331 TRS / 418 FLT
Created an example using an LLM API
Further research in prompting techniques
Week 3
(6/2/25 - 6/6/25)
Manually collected over 700 cells of 50 competitions with 14 criteria for benchmarking to ensure accurate, reliable data
Automated Collection to see if a Script utilizing an API can get close to the accuracy of the manual benchmark
Week 4
(6/9/25 - 6/13/25)
Drafted a sample Abstract
Documented a variety of data from 10 competitions from Kaggle and the results of DeepSeek-r1 performance
Script testing, using Zero-shot to test 2 different APIs resulting from training in Jetstream.
Compared and contrasted training results for areas of superiority within the specified domain of testing
Created Subsection & continued literature review for the research paper "related work" section
Preparing for Mid-Term Student Research PPT Presentation
Week 5
(6/16/25 - 6/20/25)
Worked on comparison further; compared LLM to human performance and in individual metrics
Continued literature review for related work, and found 15 good papers to reference for our research
Created charts for comparison to show in slides for Mid-Term presentation
Created Related Work slides for Mid-Term presentation
Created Related Work draft in the research paper
Week 6
(6/23/25 - 6/27/25)
Finalized Related Work
Ran an example for Keras Tuner
Finalized Zero-shot
Tested both DeepSeek-r1 & GPT-o4-mini with a total of 20 competitions each
Analyzed results and input findings in the presentation
Week 7
(6/30/25 - 7/4/25)
Worked on RAG for DeepSeek to begin testing
Configured the config.py and collection.py to utilize DeepSeek API to begin notebook collection
Debugged collection.py due to DeepSeek's updated use of "tools" syntax rather than "function calls" & tested to ensure proper output format for all 60+ competitions
Debugged consistently over the weekend, facing many challenges using RAG to generate models
Week 8
(7/7/25 - 7/11/25)
Resolved Bugs
Finished analysis of 19 competitions, waiting for finalization of 1 competition
Compared RAG & Zero-shot results vs Human
Worked on poster draft
Replaced 1 competition for all testing phases
Re-ran RAG testing for GPT-o4-mini due to data inconsistency
Worked on the poster analytics
Week 9
(7/14/25 - 7/18/25)
Poster Draft presentation
Finalize Poster
Worked on the Evaluation Draft section for the paper
Worked on the Motivation Draft section for the paper
Generated Keras Tuner DeepSeek-r1 models for testing
Started running DeepSeek-r1 models analysis
Week 10
(7/21/25 - 7/25/25)
Final week, working on research paper
Finishing Testing and analysis for Keras Tuner
Checking Deliverables
Illya and I are verifying that all data is Correct
Completed Draft of Research paper
Confirming all data is valid and saying goodbyes to the friends we made along the way.